Atera is applying generative AI to IT support through two products with deliberately different levels of autonomy. Copilot helps technical professionals investigate problems and prepare actions, while Autopilot is designed to resolve a limited set of routine issues directly for end users. Chief executive Gil Pekelman calls the approach “Action AI” because it moves beyond generating text and interacts with software and hardware. The distinction addresses a central enterprise question: when should an AI system recommend a change, and when can it be trusted to carry one out?
Action AI Connects Language to Operational Tools
Traditional chatbots can explain how to reset a password or diagnose a printer, but the user still has to follow the instructions. Atera’s model connects conversational understanding with tests and operational controls. It can examine a device or network, gather relevant evidence and determine an appropriate next step. That ability makes the system potentially more valuable than a generic assistant, while also increasing the consequences of an error.
The company’s existing platform gives the AI a unified foundation. Atera manages patching, security, alerts, tickets, automation and analytics through one codebase and database. That integration can provide context that a standalone chatbot lacks. Instead of relying entirely on a user’s description, the system can inspect technical signals. Access must be carefully governed because the same visibility that improves diagnosis can expose sensitive data or powerful administrative functions.
Copilot Keeps the Technician in Control
Copilot is intended to support IT professionals rather than execute the final fix. It can research an issue, run tests and assemble a proposed action, leaving a human to review the evidence and decide what to do. Pekelman said the product could save technicians nearly half of their working day. The claim points to the amount of time support teams spend gathering routine context before applying their expertise.
Human approval is appropriate when problems are unfamiliar, high-impact or ambiguous. A recommended network change may solve one symptom while disrupting another service. Technicians can consider business context that the model may not possess and choose a safer alternative. Copilot will be most useful if it makes its tests and reasoning visible, enabling professionals to verify the proposal rather than becoming passive approvers.
Autopilot Operates Inside Narrow Boundaries
Autopilot takes a more autonomous role for repetitive tasks such as password resets, software installation and common printer problems. Atera says it restricts the system to situations that can be handled securely and predictably. When a request falls outside those boundaries, it should escalate to Copilot and a human technician. This bounded design is a practical way to introduce agentic behaviour without granting broad authority from the beginning.
Reliable escalation is as important as successful automation. The system must recognise uncertainty, preserve a record of actions and avoid repeatedly attempting a fix that has failed. Organisations should define which devices, users and commands are eligible and ensure that permissions follow existing security policies. A fast autonomous resolution is valuable only if it does not create a larger incident.
Secure Adoption Requires Measurable Outcomes
Pekelman advised businesses to begin adopting AI while carefully selecting tools that protect data. Atera’s Microsoft collaboration is intended to keep customer information within controlled systems and prevent its use for model training. Those commitments matter in IT support, where prompts and diagnostics may contain account information, device configurations and security details. Buyers need contractual clarity and technical evidence that boundaries are enforced.
Organisations should also measure more than the number of automated tickets. Useful indicators include resolution time, recurrence, escalation quality, user satisfaction and changes made incorrectly. Comparing those outcomes with conventional support reveals whether automation genuinely improves service. It can also identify tasks that appeared routine but require more human judgement than expected.
Atera’s two-level model offers a sensible pattern for enterprise agents. AI can begin by assembling information and recommending action, then receive direct authority only for well-understood processes. As confidence grows, boundaries can expand based on observed performance rather than optimism. This approach preserves the productivity advantage of automation while recognising that operational access carries risk.
The future of AI in IT support will likely be defined by this progression from conversation to controlled action. Users do not simply want instructions when a printer fails or an account is locked; they want the problem resolved. Atera’s challenge is to deliver that convenience without making the system opaque or overpowered. If Copilot remains inspectable and Autopilot remains disciplined about escalation, Action AI could reduce repetitive work while keeping professionals responsible for the decisions that matter most.
The change will also reshape the work of support teams. Fewer routine tickets can free technicians for security, architecture and difficult user problems, but only if organisations invest in those higher-value responsibilities. Professionals will need to supervise automations, investigate failure patterns and refine boundaries as systems change. This is not the disappearance of IT support; it is a shift from manually repeating known fixes toward managing a service that includes autonomous components. Atera can help that transition by giving customers clear audit trails and controls that technicians can understand. The better the system explains what it did, the easier it becomes for teams to improve both the AI and the surrounding support process.
